Example: air traffic controller

Collaborative Knowledge Base Embedding for Recommender …

Collaborative Knowledge Base Embedding forRecommender SystemsFuzheng Zhang , Nicholas Jing Yuan , Defu Lian , Xing Xie ,Wei-Ying Ma Microsoft Research Big Data Research Center, University of Electronic Science and Technology of different recommendation techniques, Collaborative fil-tering usually suffer from limited performance due to the sparsityof user-item interactions. To address the issues, auxiliary informa-tion is usually used to boost the performance. Due to the rapidcollection of information on the web, the Knowledge base providesheterogeneous information including both structured and unstruc-tured data with different semantics, which can be consumed by var-ious applications.

network embedding method, termed as TransR, to extract items’ structural representations by considering the heterogeneity of both nodes and relationships. We apply stacked denoising auto-encoders and stacked convolutional auto-encoders, which are two types of deep learning based embedding techniques, to extract items’ tex-

Tags:

  Network, Structural, Deep, Embedding, Network embedding

Information

Domain:

Source:

Link to this page:

Please notify us if you found a problem with this document:

Other abuse

Transcription of Collaborative Knowledge Base Embedding for Recommender …

1 Collaborative Knowledge Base Embedding forRecommender SystemsFuzheng Zhang , Nicholas Jing Yuan , Defu Lian , Xing Xie ,Wei-Ying Ma Microsoft Research Big Data Research Center, University of Electronic Science and Technology of different recommendation techniques, Collaborative fil-tering usually suffer from limited performance due to the sparsityof user-item interactions. To address the issues, auxiliary informa-tion is usually used to boost the performance. Due to the rapidcollection of information on the web, the Knowledge base providesheterogeneous information including both structured and unstruc-tured data with different semantics, which can be consumed by var-ious applications.

2 In this paper, we investigate how to leveragethe heterogeneous information in a Knowledge base to improve thequality of Recommender systems. First, by exploiting the knowl-edge base, we design three components to extract items semanticrepresentations from structural content, textual content and visu-al content, respectively. To be specific, we adopt a heterogeneousnetwork Embedding method, termed as TransR, to extract items structural representations by considering the heterogeneity of bothnodes and relationships. We apply stacked denoising auto-encodersand stacked convolutional auto-encoders, which are two types ofdeep learning based Embedding techniques, to extract items tex-tual representations and visual representations, respectively.

3 Final-ly, we propose our final integrated framework, which is termed asCollaborativeKnowledge BaseEmbedding (CKE), to jointly learnthe latent representations in Collaborative filtering as well as item-s semantic representations from the Knowledge base. To evalu-ate the performance of each Embedding component as well as thewhole system, we conduct extensive experiments with two real-world datasets from different scenarios. The results reveal that ourapproaches outperform several widely adopted state-of-the-art rec-ommendation Systems, Knowledge Base Embedding , Collabo-rative Joint Learning1. INTRODUCTIONDue to the explosive growth of information, Recommender sys-tems have been playing an increasingly important role in online ser-vices.

4 Among different recommendation strategies, collaborativePermission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full cita-tion on the first page. Copyrights for components of this work owned by others thanACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-publish, to post on servers or to redistribute to lists, requires prior specific permissionand/or a fee. Request permissions from 16, August 13-17, 2016, San Francisco, CA, USAc 2016 ACM. ISBN 978-1-4503-4232-2/16/08.

5 $ : (CF) based methods, which make use of historical inter-actions or preferences, have made significant success [23]. How-ever, CF methods usually suffer from limited performance whenuser-item interactions are very sparse, which is very common forscenarios such as online shopping where the item set is extremelylarge. In addition, CF methods can not recommend new items s-ince these items have never received any feedbacks from users inthe past. To tackle these problems, hybrid Recommender system-s, which combine Collaborative filtering and auxiliary informationsuch as item content, can usually achieve better recommendationresults and have gained increasing popularity in recent years [2].

6 Over the past years, more and more semantic data are publishedfollowing the Linked Data principles1, by connecting various in-formation from different topic domains such as people, books, mu-sics, movies and geographical locations in a unified global data s-pace. These heterogeneous data, interlinked with each other, form-s a huge information resource repository called Knowledge typical Knowledge bases have been constructed, includingacademic projects such as YAGO2, NELL3, DBpedia4, and deep -Dive5, as well as commercial projects, such as Microsoft s Satori6and Google s Knowledge Graph7. Using the heterogeneous con-nected information from the Knowledge base can help to developinsights on problems which are difficult to uncover with data froma single domain [6].

7 To date, information retrieval [9], communi-ty detection [25], sentiment analysis [4] - to name a few - are thenoteworthy applications that successfully leverage the , since a Knowledge base provides rich information in-cluding both structured and unstructured data with different seman-tics, the usage of the Knowledge base within the context of hybridrecommender systems are attracting increasing attention. For ex-ample, Yu et al. [30] uses a heterogeneous information network torepresent users, items, item attributes, and the interlinked relation-ships in a Knowledge base. They extract meta-path based latentfeatures from the network structure and apply Bayesian rankingoptimization based Collaborative filtering to solve the entity recom-mendation problem.

8 Grad-Gyenge et al. [11] extended collabora-tive filtering by adopting a spreading activation based technique toincorporate a Knowledge base s network features for recommendersystems rating prediction task. However, previous studies have not1 exploited the potential of the Knowledge base since they sufferfrom the following limitations: 1) only utilize the single networkstructure information of the Knowledge base while ignore other im-portant signals such as items textual and visual information. 2)rely on heavy and tedious feature engineering process to extractfeatures from the Knowledge address the above issues, in this paper, we propose a nov-el recommendation framework to integrate Collaborative filteringwith items different semantic representations from the knowledgebase.

9 For a Knowledge base, except for the network structure infor-mation, we also consider items textual content and visual content( , movie s poster). To avoid heavy and tedious manual featureextractions, we design three Embedding components to automati-cally extract items semantic representations from the knowledgebase s structural content, textual content and visual content, re-spectively. To be specific, we first apply a network embeddingapproach to extract items structural representations by consider-ing the heterogeneity of both nodes and relationships. Next, weadopt stacked denoising auto-encoders and stacked convolutionalauto-encoders, which are two types of deep learning based embed-ding techniques, to extract items textual representations and visu-al representations, respectively.

10 Finally, to integrate collaborativefiltering with items semantic representations from the knowledgebase smoothly, we propose our final framework, which is termed asCollaborative Knowledge Base Embedding (CKE), to learn differ-ent representations in a unified model empirical studies consist of multiple parts. First, we conduc-t several experiments to evaluate the performance of three knowl-edge base Embedding components, respectively. Next, we evaluatethe effectiveness of our integrated framework by comparing withseveral competitive key contributions of this paper are summarized as the fol-lowing: To the best of our Knowledge , this is the first work leveragingstructural content, textual content and visual content from theknowledge base for Recommender systems.


Related search queries